TheSequence
·
2 weeks ago
● 4 sources
Anthropic redeployed Claude Fable 5 after a 19-day suspension for a jailbreak vulnerability, implementing a classifier that catches the exploit in over 99% of cases and downgrades requests to an older model instead of blocking them. Microsoft committed $2.5 billion and 6,000 engineers to deployment infrastructure, following similar moves by AWS ($1 billion), OpenAI, and Anthropic, signaling that model capability is now commoditized and integration has become the competitive focus. Companies are shifting from racing on model performance to building deployment systems, workbenches, and regulatory compliance layers as the primary differentiator.
The Neuron
·
2 weeks ago
Vellum launched a personal AI assistant that stores user preferences in persistent memory and handles tasks like email triage, calendar management, and meeting preparation with minimal setup required. The system processed 47 emails overnight, flagging 3 for user attention while drafting replies for the remainder. Users can deploy Vellum on multiple platforms including iOS, macOS, Web, Voice, Email, Telegram, and Slack, with options for cloud or self-hosted deployment.
The Neuron
·
2 weeks ago
● 2 sources
Claude Fable 5 achieved a 15.8% automation rate on the Remote Labor Index, a benchmark measuring how often AI agents complete real freelance projects at client-acceptable quality across 3D design, architecture, video, and other domains. The previous benchmark leader scored 4.17% eight months ago, meaning the frontier has quadrupled in less than a year, with Fable 5 roughly double the next model (Opus 4.8 at 8.3%). Human evaluators remain necessary for assessing absolute capability, as an automated AI judge overstated newer models' performance by 2–3 times despite correctly ranking them relative to each other.
